ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision", Graz, 2002 
  
  
  
  
  
  
Figure 5 — Segmentation results obtained by algorithm 2 on a 
test area of 27-27 m. TopoSys data set. 
TESTS 
A 5-dimensional case 
A series of tests in the n-dimesional case was performed using 
a laser data set provided by FOTONOR AS for the ISPRS test 
on estracting DEMs. The test area was scanned with an Optech 
laser scanner, and both first and last pulse data were recorded. 
The sampling density is of 1.8 points per square meter. The 
scanner recorded the received pulse intensity also, and we have 
used this data with descriptor mapping to improve the 
segmentation results. In this test we have mapped the total static 
moment only. The total dimension of the problem is: 
Sao'up 
and the best robustness reached using five dimensions, supplies 
the low sampling density. 
      
. 
^ 
I 
. *4 
Mut 
N . ; . nva ; : : 
HE 4 A.) ; À 
uen Canet t. Yo,» Rm, .. . 
A Se te fis D an . * s m LE 
oo ated tena PAPA ee "s e [7 . "a 5 
* wien, weve vent pe ALME aera, “yo, 
A 
. 
PN posts wc m —Ü lá : Medo st 
2 A 
uS roue, " 
——Mg sam ml n 
Ift vet — pan, 
Figure 6 — Segmentation results obtained on a test area of 64:52 
m working in three, four and five dimensions. The black line 
marks the junctions and the edges in which the algorithm fails, 
because a tree touch the roof. 
Let's consider three cases. 
Case 1: we have segmented a test area of 64:52 m, working 
at first in the three dimensions X, Y and Z only and using 
the algorithm 2. The algorithm fails in the junctions between 
different entities. 
Case 2: working in the four dimensions of the space of the 
objects (X, Y, Z and radiometry) the algorithm 2 don't fails 
in the junctions. 
Case 3: we have used the algorithm 1 and the total static 
moment as descriptor, working in five dimensions. The 
algorithm don't fails in the junctions and the result is more 
precise on the edges. 
The results are in figure 6. 
Test on a large data set. 
We have performed a test on a large data set (about 500000 
points) of laser measuraments, scanned on urban area. We 
have used algorithm 1 working in five dimensions, as in the 
case 3 of the previous test, and we want to look at the time 
of work of the algorithm. The processor used is an AMD 
Athlon with CPU clock of 900 MHz. The result of the test is 
satisfying, for the segmentation output (figure 7) and for the 
short time of work (only about ten minutes!). 
  
  
  
  
  
Figure 7 — Segmentation results obtained on a large data set, 
covering an urban area of 422-634 m. Note ground, gruped 
in a unique entity. The entities with less than 50 points are 
in white. 
A - 293